Grok 3
xAI's first frontier reasoning model: Think mode plus DeepSearch, trained on Colossus.
About
Announced February 17, 2025, Grok 3 was xAI's first serious frontier model, trained with roughly 10x the compute of Grok 2 on the Colossus supercomputer (reportedly 200,000+ NVIDIA H100 GPUs). It introduced two signature modes: Think, a visible chain-of-thought reasoning mode for math, science, and code, and DeepSearch, an agentic web-and-X research mode that builds cited reports. Variants include Grok 3 mini and a reasoning-focused Grok 3 (Think). xAI claimed it matched GPT-4o-class models at launch and beat o3-mini on AIME 2025 math. It carries xAI's looser, more opinionated default personality versus competitors.\n\nLimitations: access is paywalled behind X Premium+ ($40/mo at launch) or SuperGrok ($30/mo); free windows were time-limited. Early versions drew criticism over weaker guardrails and misinformation risk. It was quickly followed by Grok 4 (July 2025), making Grok 3 a transitional release.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
xAI-reported: beat o3-mini on AIME 2025; competitive with GPT-4o at launch
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 50; Trending score 82/100 from researcher assessment.